PyRaDiSe

PyRaDiSe: DICOM-based Radiotherapy Data Processing and Deep Learning Integration

PyRaDiSe processes DICOM data for radiotherapy, enabling automated handling of image series, DICOM-RT Structure Sets, and registrations, and supporting integration of deep learning models for medical image segmentation.

Key Features:

• DICOM-RT Structure Set Handling: Processes DICOM RT Structure Sets using 2D-based and 3D-based conversion methods to generate high-quality contours while minimizing pixelation artifacts.
• Framework-Agnostic Deep Learning Integration: Integrates with PyTorch and TensorFlow for model inference without framework dependency.
• Automated DICOM Processing: Automates processing of DICOM image series, structure sets, and spatial registrations.
• Partially Invertible Pre- and Post-processing: Applies processing routines that preserve spatial properties, including image origin and orientation, to maintain segmentation accuracy.
• Dataset Construction: Supports generation of structured datasets for training deep learning segmentation models.

Scientific Applications:

• Auto-Segmentation in Radiotherapy: Enables organ-at-risk segmentation in brain tumor patients and facilitates deployment of deep learning models in clinical radiotherapy workflows.
• Radiotherapy Dataset Preparation: Standardizes DICOM data for training and validation of medical image segmentation models.

Methodology:

Implements structured DICOM data management with automated preprocessing and postprocessing pipelines that maintain spatial metadata. Converts DICOM-RT Structure Sets using 2D and 3D methods, integrates deep learning model inference via PyTorch or TensorFlow, and preserves geometric consistency throughout segmentation and dataset generation workflows.